Malte Schilling

… to the homepage of Malte Schilling. On this site, you can read about my current work (see my research statement ), latest publications …

I am a Responsible Investigator at the Center of Excellence for ‘Cognitive Interaction Technology’ in Bielefeld. My work concentrates on internal models, their grounding in behavior and their application in higher-level cognitive function like planning ahead or communication.

In general, my interest is on understanding cognitive mechanisms and how to leverage these in real systems. Therefore, I am following a minimal cognitive systems approach that is applying such mechanisms in real robotic systems. In particular, I started to work on adaptive systems for the control of six-legged walking robots. This perspective is, on the one hand, inspired on the decentralized organization of biological systems and, on the other hand, on the recruitment of grounded internal models and their flexible application in internal simulation. The systems are realized as recurrent neural networks and my current focus is, first, on learning novel behaviors using a (decentralized deep) reinforcement learning approach that exploits detailed information on the modular structure of biological control systems. Second, extending the idea of grounded internal models towards transferring learnt models for different cognitive tasks (e.g. planning or observation) and exploiting local information on joint dynamics in a latent dynamic space. In the long run, my goal is understanding and constructing a cognitive system that is adaptive as well as can deal with novel tasks, learn novel solutions and, furthermore, can interact with other cognitive systems.

Before coming back to Bielefeld I stayed for two years in Berkeley (California) at the ICSI (International Computer Science Institute) as a PostDoc and did research on the connection of higher-level linguistic representation to sensorimotor representation as part of the Neural Theory of Language project. During a stay at the Sony Computer Science Laboratory in Paris I worked on the embodiment of language and the connections between internal models and language. Recently, I went for a brief stay back to Sony CSL in Tokyo and worked on grounding of latent dynamic models.

I finished my PhD in January 2010. It is on a bottom-up approach to cognitive control. The thesis was written in the working group of Holk Cruse (Biological Cybernetics and Theoretical Biology) and deals on the one hand with the implementation of a biology-inspired control system for a hexapod robot. On the other hand, this system has been extended through grounded internal models which are used by the lower-level system for control of behaviour, but can now be additionally applied in a mental simulation for planning ahead. The internal body model has been realised as a neural network representing the body structure and its dynamics in a dual quaternion representation using the Mean of Multiple Computation principle (for details, see publications).